Google Translate

Those outside the translation industry may think that Google Translate is the answer to every translation need. It is free, its abilities are constantly growing thanks to crowd sourcing and it is available in numerous languages. Plus its new camera tool and downloadable dictionaries via the mobile app allow it to function everywhere. What’s not to love?

But is it really as good as a human brain? While the above points are all true, Google Translate (and machine translation in general) has its limits.

Having recently stumbled across the BMJ article “Use of Google Translate in medical communication: evaluation of accuracy” by Patil and Davies (BMJ 2014;349:g7392), it is interesting to note that research has been carried out on this topic. Anecdotally one often hears stories about a bad translation via Google Translate, but to have statistics available to vouch for its success (or lack of) makes for an interesting read.

The research involved translating ten commonly used medical phrases from English into 26 languages using Google Translate. After evaluation by human translators, it was found that of the total 260 translated phrases 150 (57.7%) were correct and 110 (42.3%) showed errors, some very serious. Of the combinations translated, Western European languages fared best under testing with an accuracy rate of 74% (with Portuguese as high as 90%) whilst African languages showed the most errors (particularly Swahili, with only 10% of phrases translating correctly).

As referred to in the original article, some of these errors included the translation of “Your child is fitting” from English into “Your child is dead” in Swahili and “Your husband has the opportunity to donate his organs” from English into “Your husband can donate his tools” in Polish. Such fallacies could cause untold worry and angst for families in the medical setting.

Whilst breaking the language barrier between doctors and patients is an essential step in communication, when it comes to important discussions and reports – especially those to be analysed in the context of clinical trials – there can be no replacement for a skilled linguist. The article itself states “Google Translate should not be used for taking consent for surgery, procedures, or research from patients or relatives unless all avenues to find a human translator have been exhausted, and the procedure is clinically urgent”.

However, that said, as indicated by a Dutch responder to the original article, with careful management regarding the inputting of phrases by someone with a little linguistic knowledge of the target language and a good awareness of how Google Translate’s algorithms function, results could be improved. Unfortunately, non-linguist users of this tool do not always have these skills.

Nevertheless, there are many Computer Assisted Translation (CAT) tools available for skilled translators to use on a daily basis. Translation memories save segments of a translator’s previous work allowing these phrases to be re-used in future translations. Once such software has been mastered, it not only helps to speed up content production in the long run, but it importantly assists with maintaining linguistic consistency. Separate memories can be saved by translators for each of their clients, meaning that a client’s preferred terminology can be used in each new translation. Terminology tools and termbases can be employed alongside such translation memories to store approved terms and usage criteria. However, it is imperative to note here that the computer alone is not doing the work for the translator, but aiding the translator on the path to a successful human translation.

With written patient reports and clinical trials documentation, Parallel always uses human translators to carry out its translations, some of whom may benefit from the assistance of CAT tools. However, the key is that our translators proof their own work and upon submission, these translations are then checked twice (by other humans) in-house at Parallel to ensure accuracy and quality compared to the source text. When it comes to medical and pharmaceutical matters, particularly those of life or death, it seems that the Google Translate error rate discovered by this research is unacceptable.

Perhaps it’s best to stick to Google Translate for your summer holidays after all.

Reference: BMJ 2014;349:g7392 (